AML Analytics: A journey from the Quantitative to the Qualitative Model
All banks and financial institutions are now required to have a framework for combatting money laundering that will identify suspicious transactions and report them to the appropriate regulatory bodies.
Challenges posed by the framework — During and after implementation
Generation of a large number of false positives
This the most common challenge faced. Transaction-monitoring platforms currently in use have quantitative models at their core. However, an implemented model generates large volumes of alerts that need to be reviewed by operations teams. Close to 98% of these are false alerts, according to industry estimates. The challenges they pose are due to the following reasons:
1. Increased regulatory oversight and scrutiny
2. Constant changes to technology
3. Ever-expanding business and the resulting new anti-money laundering (AML) typologies (i.e., new methods, techniques and trends of money laundering)
4. Frequent changes to the configurations of quantitative models
5. Increasing need for skilled resources, with both technical and business skills
6. The need to build and train large teams for review and remediation
However, financial institutions face a conundrum: although they may have the best-in-class AML framework in place, they may still be fined by regulators for a lack of sound, adequate controls and reporting.
Improper implementation of controls
Controls are selected based on an entity’s risk profile, and the risk identified by analyzing the risk profile is a collation of risk areas and red flags that need to be addressed. However, many of a business’s activities may remain unmonitored, for the following two reasons:
1. As financial institutions expand into new geographies, products and channels, and as AML typographies change
2. Due to gaps between the risk identified by analyzing the risk profile and the controls implemented. This could be because of incorrect configurations of the model implemented, thresholds and scoring values, or mathematical calculations.
Every change in business or regulatory requirement has necessitated adding data points and defining new workflows. The approach has been to add patches to existing core systems (as opposed to a complete overhaul of the system). This has resulted in data issues, and incomplete, inaccurate, and insufficient data cleansing and treatment.
Current methods of implementation mostly focus on thresholds and scoring values implemented at the organizational or legal entity level, based on the assumption that ‘one size fits all’. However, customer behavior is a combination of geography, product, channels and customer profile. We understand, though, that applying thresholds and scoring values at the customer level is not feasible; therefore, we believe similar transactional behaviors need to be identified and addressed as segments. This would help to delocalize the configurational settings and make the implemented controls more robust.
Challenges posed by regulation and the need for skilled resources
Money laundering reporting officers (MLROs) or chief compliance officers (CCOs) have generally been thought of as functional experts dealing with this particular domain. However, as the field of compliance has advanced, these roles have expanded to include many more responsibilities. The responsibility for all control implementations and the technical ownership of the implemented controls now lie with these professionals. This requires that they have both domain and technical knowhow. This is, therefore, a niche skillset and hard to find.
We frequently hear that ‘machine learning and artificial intelligence’ is the solution — the qualitative approach, as opposed to the popular quantitative models. However, before implementing either of these, we need to focus on the underlying data. Data is key for qualitative models, and these models are very sensitive to the data being fed. The data needs to be cleansed, sliced and diced, according to a model’s requirements, and the correct skillset is required to understand the data and the treatment required for a particular type of data. We believe that qualitative models based on machine learning and artificial intelligence could be used as a platform to solve problems relating to segmentation, threshold tuning and gap analysis.
How we can help you
Acuity Knowledge Partners’ methodology and roadmap help financial institutions augment their efforts to address the challenges on both the people and technology fronts. Our team of business experts and data scientists can help you by
1. Supplementing resources to perform risk mapping, before and after implementation; model validation; threshold tuning; and segmentation
2. Providing technology solutions to increase the frequency of AML analytics (quantitative models)
3. Partnering to define, build and manage qualitative models
Orignal source : https://www.acuitykp.com/